Related Experiment Videos
Reproducibility and accuracy of interactive segmentation procedures for image analysis in cytology
A J Einstein1, J Gil, S Wallenstein
1Department of Biomathematical Sciences, Mount Sinai School of Medicine, New York, NY 10029, USA. einstein@msvax.mssm.edu
Journal of Microscopy
|January 7, 1998
Summary
The arc-fitting method offers the most reproducible nuclear image segmentation for cancer diagnosis. Reproducibility was high across methods, but arc-fitting excelled, impacting accuracy.
Area of Science:
- Medical image analysis
- Computational pathology
- Cancer diagnostics
Background:
- Accurate nuclear image segmentation is vital for cancer diagnosis using image analysis.
- Evaluating segmentation method reproducibility and accuracy is essential for clinical application.
Purpose of the Study:
- To assess the reproducibility and accuracy of interactive nuclear image segmentation methods.
- To compare thresholding, manual tracing, arc-fitting, and ellipse-fitting routines.
Main Methods:
- Segmentation of nuclei using four interactive methods.
- Derivation of nuclear size and shape features.
- Statistical analysis using variance component models and intraclass correlation coefficients.
Main Results:
- Arc-fitting demonstrated the highest reproducibility; thresholding showed the least.
- High intraobserver and interobserver reproducibility was observed across methods.
- Measurement accuracy, particularly for area, correlated with method reproducibility.
Conclusions:
- Arc-fitting is the most reproducible segmentation method for nuclear images in cytological diagnosis.
- Segmentation method choice impacts reproducibility and accuracy, with arc-fitting being superior.
- Sample size is more influenced by tissue variability than segmentation method or observer number.